Discussion Intel current and future Lakes & Rapids thread

[H433x0n]

Oh? What "SRF session Q&A" would that be?

He’s probably talking about hot chips.

Technically Crestmont is based off of Gracemont. The Sierra Glen core is literally called “CME”, where Crestmont is “CMT”. Sierra Glen (CME) is a die shrink of Crestmont as far as I know (5 wide re-allocate though).

Crestmont is essentially Gracemont that went 6-wide and has a 6-8% IPC bump. It’s basically splitting hairs.

 

[Henry swagger]

https://twitter.com/x/status/1704168875229954397

sierra forrest 😊 gor cameras

 

[poke01]

Nice surprise

 

[JoeRambo]

Is there anything wrong with 288 Gracemonts at some 2.5Ghz ? Not really, it is amazing amount of compute for this same niche that Z4C, Ampere are targeting. In fact 144C with the right price was already enough. AMD will respond with 192C Z4C or Z5C and competition will heat up -> whomever looses, the customers win.

Sounds like perfectly fine product for me, 5-10% more IPC from improved design is also perfectly fine and won't change much. If some resident vendor warriors want to claim that people will choose 128C Ampere or Z4C due to per core IPC ( while at same time ignoring much stronger offering of 96C Z4X3D ) -> all power to them.

In 2025 Intel plans to release another "sea of atoms" product on more advanced process, so those who will be unhappy with 1st gen TCO, might become more happy with 2nd gen.

Overall i like what Intel did with Atom here: having 2 or 4 cores pool their L2 is the correct move in this segment and enables additional areal density, something competitors can't match.
And i would not be surprised if they did the same with P-Cores, imagine having 6+MB of L2 to play with at the cost of no additional area?

 

[DrMrLordX]

Is there anything wrong with 288 Gracemonts at some 2.5Ghz ?

There will be if Intel never produces a meaningful number of these chips.

 

[JoeRambo]

There will be if Intel never produces a meaningful number of these chips.

Should their customers really care about this? If the price is right, orders are fullfilled above or below table, they will not care.

What's next, caring about Intel's yields and profits while producing palm sized "chiplets" ? Let's leave those concerns to Intel's bean counters and stick to technical.

 

[dullard]

I think you got the grammar wrong. "It" would be referring to the subject of the previous sentence.

Two rules of pronouns would be broken in Abwx's post.

1) Pronoun antecedents should never be ambiguous. Everyone should instantly know what "it" refers to. Consider this sentence: "After removing the CPU from the motherboard, Joe NYC sold it." What did you sell? Will everyone agree on what you sold? Probably not. The "it" could be the CPU. The "it" could be the motherboard. No one knows exactly what you sold from that sentence. Thus, the Joe NYC selling sentence broke the ambiguous antecedent rule. That is the main problem we are talking about here. There were multiple possible processes that could have been implied in Abwx's post.

2) The pronoun's antecedent should be close by. The antecedent should never be in a previous chapter, previous page, or even a previous paragraph. An antecedent could be in a previous sentence, but only if clear. Look up far away antecedents. Here is one reference: https://www.trentu.ca/academicskills/how-guides/how-edit-your-writing/grammar-and-style/pronoun-agreement-and-reference#:~:text=Every pronoun should refer to,noun the pronoun refers to.

Every pronoun should refer to a specific antecedent that has been mentioned and is close by. If the antecedent is missing or too far away from the pronoun, it can be difficult for the reader to understand what noun the pronoun refers to.

Or another example: https://www.dummies.com/article/aca.../positioning-pronoun-antecedent-pairs-190485/

One way to lose a reader is to let your pronouns wander far from their antecedents. To avoid confusion, keep a pronoun and its antecedent near each other.

Rule #1 was broken; there were multiple antecedents. Anyone who applies rule #2 of being close by will be mislead since his antecedent was both the furthest away and also in a different paragraph. Being the subject of a sentence has nothing to do with properly using pronouns.

 

[dullard]

He’s probably talking about hot chips.

Technically Crestmont is based off of Gracemont. The Sierra Glen core is literally called “CME”, where Crestmont is “CMT”. Sierra Glen (CME) is a die shrink of Crestmont as far as I know (5 wide re-allocate though).

Crestmont is essentially Gracemont that went 6-wide and has a 6-8% IPC bump. It’s basically splitting hairs.

Here is Anandtech's coverage of Sierra Forest in Hot Chips 2023:

AnandTech Forums: Technology, Hardware, Software, and Deals

Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

www.anandtech.com

H433x0n is correct Crestmont and Gracement are different, but not much different.

So Granite and Sierra can be thought of as a deconstructed Meteor Lake processor, with Granite getting the Redwood Cove P-cores, while Sierra gets the Crestmont E-Cores.
...Intel is revealing that Crestmont is offering a 6-wide instruction decode pathway as well as an 8-wide retirement backend...Meanwhile, new to the E-core lineup with Crestmont, the cores can either be packaged into 2 or 4 core clusters, unlike Gracemont today, which is only available as a 4 core cluster.

 

[mikk]

.@intel will first test high-NA EUV from @asml with the "18A" process node and use high-NA for the production of a "Intel next" node, tho no real timeline was given (I'd say after 2025). But it does look like Intel will be first on high-NA, after they were years behind with EUV.

https://twitter.com/x/status/1704440799163732233

No High-NA for 18A. Nothing new really but some people thought it's coming.

 

[Geddagod]

https://twitter.com/x/status/1704440799163732233

No High-NA for 18A. Nothing new really but some people thought it's coming.

Ye, we talked about this a couple months ago. The head of Intel's fab development (Kellher? IIRC) said that Intel 18A is able to use high NA EUV, but in order to reduce risk, they are also able to develop Intel 18A without High NA EUV. I'm pretty sure the default was that it would use high NA, but it looks like due to delays or cost they won't end up using it as originally planned.

 

[Geddagod]

Here is Anandtech's coverage of Sierra Forest in Hot Chips 2023:

AnandTech Forums: Technology, Hardware, Software, and Deals

Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

www.anandtech.com

H433x0n is correct Crestmont and Gracement are different, but not much different.

View attachment 86010

IPC increase was quoted to be 6-8% IIRC (read it somewhere but don't remember where lol). Doesn't sound too bad. A bit surprising that SRF isn't using crestmont, but it looks like the all core frequency reduction is greater than the IPC gain?

 

[dullard]

Ye, we talked about this a couple months ago. The head of Intel's fab development (Kellher? IIRC) said that Intel 18A is able to use high NA EUV, but in order to reduce risk, they are also able to develop Intel 18A without High NA EUV. I'm pretty sure the default was that it would use high NA, but it looks like due to delays or cost they won't end up using it as originally planned.

18A was originally going to be timed exactly with high NA EUV.

But when Intel moved 18A from H2 2025 to H2 2024, that meant 18A would come out before high NA EUV.

There are rumors that high NA EUV was delayed. But, I think the real change is Intel's rush to get so many foundry customers onto 18A as soon as possible.

 

[aigomorla]

Two rules of pronouns would be broken in Abwx's post.

1) Pronoun antecedents should never be ambiguous. Everyone should instantly know what "it" refers to. Consider this sentence: "After removing the CPU from the motherboard, Joe NYC sold it." What did you sell? Will everyone agree on what you sold? Probably not. The "it" could be the CPU. The "it" could be the motherboard. No one knows exactly what you sold from that sentence. Thus, the Joe NYC selling sentence broke the ambiguous antecedent rule. That is the main problem we are talking about here. There were multiple possible processes that could have been implied in Abwx's post.

2) The pronoun's antecedent should be close by. The antecedent should never be in a previous chapter, previous page, or even a previous paragraph. An antecedent could be in a previous sentence, but only if clear. Look up far away antecedents. Here is one reference: https://www.trentu.ca/academicskills/how-guides/how-edit-your-writing/grammar-and-style/pronoun-agreement-and-reference#:~:text=Every pronoun should refer to,noun the pronoun refers to.

Or another example: https://www.dummies.com/article/aca.../positioning-pronoun-antecedent-pairs-190485/

Rule #1 was broken; there were multiple antecedents. Anyone who applies rule #2 of being close by will be mislead since his antecedent was both the furthest away and also in a different paragraph. Being the subject of a sentence has nothing to do with properly using pronouns.

I think this is a bit too much over the top in saying what the post implied was ambiguous. We are a technical forum, not a English 101, and to many English is a second / third / fourth language.

In the future, i think a " "IT" was ambiguous and could of represented many other things except what was implied" is enough then the wall of text teaching the connotation of the word IT.

Moderator Aigo

 

[JoeRambo]

Crestmont is essentially Gracemont that went 6-wide and has a 6-8% IPC bump. It’s basically splitting hairs.

I think most interesting thing is that Intel lately is choosing to deliberately underplay it's hand, we already saw that with 144C => 288C. It's the same with Crestmont, just a month or so ago, they were claiming it is 5 wide allocate

https://images.anandtech.com/doci/20034/HotChips%202023%20Press%20Briefing%20Final__08.png

Nowadays it seems they have found a way to make it 6 wide in allocate?





I think Intel is reading forums and Twitters and spreading FUD on purpose and having good fun.

Their rename / allocate / was 2010's effort before, who knows maybe they have found someone else has good design in the same company and it's no longer mobile phone worthy?


( source like always, golden material from Chips and Cheese, all credit to them )

 

[Geddagod]

It's the same with Crestmont, just a month or so ago, they were claiming it is 5 wide allocate

https://images.anandtech.com/doci/20034/HotChips%202023%20Press%20Briefing%20Final__08.png
Nowadays it seems they have found a way to make it 6 wide in allocate?

No, SRG is rumored to be ported gracemont, so you get the 5 wide rename but higher clocks iso power (because narrower architecture) , crestmont is wider. SRG was at first thought to be ported crestmont, but nah, it's ported gracemont

 

[dullard]

No, SRG is rumored to be ported gracemont, so you get the 5 wide rename but higher clocks iso power (because narrower architecture) , crestmont is wider. SRG was at first thought to be ported crestmont, but nah, it's ported gracemont

I just checked and just about every tech news site says Crestmont. Heck, all 3 of 3 of Anandtech's articles on the subject say it is not Gracemont. https://www.anandtech.com/tag/sierra-forest

Do you have anything that verifies it is Gracemont?

 

[mikk]

IPC increase was quoted to be 6-8% IIRC (read it somewhere but don't remember where lol). Doesn't sound too bad. A bit surprising that SRF isn't using crestmont, but it looks like the all core frequency reduction is greater than the IPC gain?

4-6% reported by multiple sites, apparently this is what Intel told them. This modest improvement is more than I expected from Crestmont though. I thought Redwood Cove brings more and now it's the other way around Crestmont brings more IPC impovements than Redwood Cove? This was unexpected.

 

[Saylick]

Hm, so based on what C&C reported for IPC, it seems like Crestmont is somewhere between Zen 2 and Zen 3 in IPC, but leaning more towards Zen 2.

 

[DrMrLordX]

Should their customers really care about this? If the price is right, orders are fullfilled above or below table, they will not care.

What's next, caring about Intel's yields and profits while producing palm sized "chiplets" ? Let's leave those concerns to Intel's bean counters and stick to technical.

Stick to technical? Okay so you're an ODM and Intel says "we cancelled Sierra Forest-AP, so don't worry about filling orders for those" and then at a meeting Intel is all "lol we totally are producing Sierra Forest-AP". So the technical angle is: where there should be a server CPU, instead you get airy nothing. So how well does an empty promise perform? What is the perf/W on that?

 

[RTX2080]

SRF-AP is a thing which is off-roadmap. That said please don't put too much hope on it.

 

[Geddagod]

I just checked and just about every tech news site says Crestmont. Heck, all 3 of 3 of Anandtech's articles on the subject say it is not Gracemont. https://www.anandtech.com/tag/sierra-forest

Do you have anything that verifies it is Gracemont?

Hotchips when Intel talked about Sierra Glenn. They claimed it has a 5 wide rename, not 6 like crestmont has.

 

[Geddagod]

4-6% reported by multiple sites, apparently this is what Intel told them. This modest improvement is more than I expected from Crestmont though. I thought Redwood Cove brings more and now it's the other way around Crestmont brings more IPC impovements than Redwood Cove? This was unexpected.

Ah that's my bad. I guess I misremembered the exact range of values.

 

[inf64]

Ah that's my bad. I guess I misremembered the exact range of values.

Intel lists IPC upgrades only for E cores. For P cores they state "improved performance efficiency"




It's safe to say that P cores are on the same IPC level as Raptor Lake, which is not bad. The bad part is the clock regression vs top mobile Raptor Lake parts.

 

[dullard]

Hotchips when Intel talked about Sierra Glenn. They claimed it has a 5 wide rename, not 6 like crestmont has.

That must have been someone misspeaking the number 5 instead of 6. Every major site that covered Sierra Glen at Hot Chips says it is Crestmont and not Gracemont. Intel even shared many slides listing the new Sierra Glen E-core features at Hot Chips. https://www.intel.com/content/www/u...-forest-xeon-press-briefing-presentation.html

 

[Henry swagger]

https://twitter.com/x/status/1704765286400798739

Gracemont has impressive ipc for a small core.. crestmont has 8% plus ipc.. Sierra forrest and clearwater should dominate with ease😏🏅💻